[ntvcojp] Extract NUXT data (#1915)

Fixes: https://github.com/ytdl-org/youtube-dl/issues/30309
Authored by: nao20010128nao
This commit is contained in:
The Hatsune Daishi 2021-12-08 02:03:48 +09:00 committed by GitHub
parent 443b21dc4e
commit ddd24c9949
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

View File

@ -3,8 +3,9 @@ from __future__ import unicode_literals
from .common import InfoExtractor
from ..utils import (
js_to_json,
ExtractorError,
smuggle_url,
traverse_obj,
)
@ -19,7 +20,7 @@ class NTVCoJpCUIE(InfoExtractor):
'ext': 'mp4',
'title': '桜エビと炒り卵がポイント! 「中華風 エビチリおにぎり」──『美虎』五十嵐美幸',
'upload_date': '20181213',
'description': 'md5:211b52f4fd60f3e0e72b68b0c6ba52a9',
'description': 'md5:1985b51a9abc285df0104d982a325f2a',
'uploader_id': '3855502814001',
'timestamp': 1544669941,
},
@ -28,22 +29,30 @@ class NTVCoJpCUIE(InfoExtractor):
'skip_download': True,
},
}
BRIGHTCOVE_URL_TEMPLATE = 'http://players.brightcove.net/%s/default_default/index.html?videoId=%s'
def _real_extract(self, url):
display_id = self._match_id(url)
webpage = self._download_webpage(url, display_id)
player_config = self._parse_json(self._search_regex(
r'(?s)PLAYER_CONFIG\s*=\s*({.+?})',
webpage, 'player config'), display_id, js_to_json)
video_id = player_config['videoId']
account_id = player_config.get('account') or '3855502814001'
player_config = self._search_nuxt_data(webpage, display_id)
video_id = traverse_obj(player_config, ('movie', 'video_id'))
if not video_id:
raise ExtractorError('Failed to extract video ID for Brightcove')
account_id = traverse_obj(player_config, ('player', 'account')) or '3855502814001'
title = traverse_obj(player_config, ('movie', 'name'))
if not title:
og_title = self._og_search_title(webpage, fatal=False) or traverse_obj(player_config, ('player', 'title'))
if og_title:
title = og_title.split('(', 1)[0].strip()
description = (traverse_obj(player_config, ('movie', 'description'))
or self._html_search_meta(['description', 'og:description'], webpage))
return {
'_type': 'url_transparent',
'id': video_id,
'display_id': display_id,
'title': self._search_regex(r'<h1[^>]+class="title"[^>]*>([^<]+)', webpage, 'title').strip(),
'description': self._html_search_meta(['description', 'og:description'], webpage),
'title': title,
'description': description,
'url': smuggle_url(self.BRIGHTCOVE_URL_TEMPLATE % (account_id, video_id), {'geo_countries': ['JP']}),
'ie_key': 'BrightcoveNew',
}